{{ title }}

Ihave been turning the same thought over for a long time, and I am writing it down mostly to find out whether it holds.

It begins somewhere most of us never think about. It is a little past three in the morning, it is cold, and someone is standing alone in a barn with an animal that is dying. Beside them is a farmer who has put most of their livelihood, and a fair amount of their sleep, into those animals, and who is watching. The person who has to decide what happens next has no colleague in the corridor to lean toward, no one to glance at the case and say yes, that looks right. The decision is theirs alone, it has to be made now, and no one is going to check it afterward.

I should say plainly, before going further, that I am not a veterinarian. I have never stood in that barn at three in the morning, and I never will. I am the founder of the project and platform I am going to describe, not the clinician at the center of it. I came at this as a builder.. an outsider who kept noticing a gap and could not let it go, and everything clinical in what follows belongs to the veterinarians and academics I work with, not to me. I think that distance is part of why I can see the problem clearly, and it is certainly why the whole thing is built so that people with the actual expertise, not me, have the final word.

I have become more and more preoccupied with that exact situation, not because it is rare, but because it exposes a mistake I think we make whenever we talk about knowledge. We pretend the difficulty for that person in the barn is that they do not know enough: that if they had only read more, remembered more, kept more facts within reach, the night would be easier. But that is not what I see, from the outside looking in. What I see is something more uncomfortable, and more human. It is the loneliness of carrying a responsibility no one shares.

Judgment was always something we did together

We have a habit of imagining expertise as something that sits inside a single head. The capable professional, alone with their knowledge. But if you look closely at how clinical judgment actually forms and stays sharp, you find it has almost never been a private matter. It was maintained collectively. The ward round where several people walked bed to bed together. The colleague you grabbed for thirty seconds by the coffee machine. The second opinion you asked for without making a thing of it. The quiet, constant calibration against other people who know the same craft.

The philosopher Michael Polanyi gave a name to why this matters so much. He wrote that we know more than we can tell, that a great deal of real competence is tacit, held in the hands and the gut, and stubbornly resistant to being written down (infed.org). Tacit knowledge transfers badly through textbooks. It transfers master to apprentice, by standing next to someone, by watching where an experienced person hesitates precisely where there is reason to hesitate. Remove the community and you do not only remove the comfort of company. You remove the very mechanism that kept the judgment honest.

And that is exactly what happens when a profession is stretched too thin across the map. The person alone in the barn at night is not only lonely in the human sense. They are cut off from the collective process the craft has always quietly depended on.

What the profession knows but rarely says out loud

Here I have to be specific, because the numbers are harder than most people realize. Veterinarians die by suicide far more often than the general population. A careful study found male veterinarians 2.1 times and female veterinarians 3.5 times as likely to die by suicide as everyone else (JAVMA). A recent cross-national study that includes Norway describes the same alarming pattern of psychological distress in the profession (Frontiers). One veterinarian summarized the weight of it in an interview with a sentence I have not been able to forget.

You are always going to be failing somebody.

The causes are tangled, and I want to be careful not to make them simpler than they are. Long hours, financial pressure, demanding owners, the nearness of euthanasia and death. But one thread keeps returning, and it is the one I have been circling. The feeling of standing alone with decisions that mean everything, without a community to lean on.

In Norway this carries an extra dimension, because it is also a question of preparedness. The government itself has examined access to veterinary services across the country and pointed out how fragile coverage is in the rural districts (regjeringen.no). Professional voices describe the shortage of district veterinarians as a threat to food-supply preparedness itself (VetNorden). When a critical profession thins out geographically, it is not only hands that disappear. A network disappears, and with the network goes the silent calibration that lets a young veterinarian trust their own judgment at three in the morning.

A landscape with almost nothing in it

When I first went looking, as a layperson trying to map a field that was not my own, I expected the problem to be that the knowledge was scattered. It was worse than that. For Norwegian production-animal medicine, the cattle and sheep and pigs and poultry that feed the country, there is no equivalent of the out-of-hours handbook that human medicine takes for granted. Therapeutic recommendations are thin and fragmented. International material is of limited use, because disease patterns, regulations, and approved treatments are local. A veterinarian in a Norwegian barn cannot simply read what an American database says about the same species and apply it. The very thing that would help most, a consolidated, practical, Norwegian body of clinical guidance, mostly did not exist in one place.

There is a clean comparison in human medicine, and it taught me how to think about this. Norwegian doctors pay for a clinical decision-support tool called NEL, and at first that seems strange, since doctors are confident people who believe they know their field. But NEL is not sold as a textbook at all. It is sold as three other things entirely. Decision support at the edges, for the rare case no one sees often. Quality assurance and a defensible record: a way to check your own judgment against current guidelines, which matters enormously because the law holds you to a standard of soundness. And ready patient information to hand over. The business that runs NEL reached around 82 million kroner in revenue at a 41 percent operating margin (NHO Trøndelag). Nobody builds margins like that selling a book. They are selling confidence and a backstop.

The veterinary world has the same drivers, only sharper. The standard of professional soundness exists there too, under the animal-health-personnel law and the oversight of the food safety authority. The time pressure is identical, and worse out in the field. And the switching between domains is brutal, since a production-animal vet can move between cattle, sheep, pigs, and poultry in a single day. So the thing to build was never a textbook. It was clinical decision support with a backstop. That distinction changed everything about how I approached it.

The part that was genuinely hard to build

I want to be honest about the engineering, because it is easy to make this sound simple in hindsight, and it was not. Building a knowledge system that a clinician can trust at three in the morning is one of the harder things I have tried to do, and most of the difficulty hides in places a casual observer would never look. Three problems, in particular, took far more than I expected.

That refusal is built, not wished for. The system leans heavily on retrieval, so that every clinical claim is tied back to a specific source rather than spun from the model’s own confidence. The generation step is fenced in by the retrieved material, the prompts are written to make the model abstain rather than improvise when the source is thin, and anything that touches treatment carries its citation with it the way a good clinician carries their reasoning. The aim is that the tool would rather say I do not have a sound basis for this than produce a fluent guess.

None of these are problems you solve by being clever once. They are problems you solve by building a careful pipeline and then maintaining the discipline to keep it honest. And expert approval is a required step, not an optional courtesy. This last point matters most to me precisely because I am not the expert. I cannot judge whether a treatment recommendation is sound, and I have built the system on the assumption that I never should. Practitioners and academics sign off on what goes in. My job is to make their judgment easier to capture, safer to store, and faster to reach, not to substitute my own for it. Underneath all of it sits a stack chosen so that a very small team can run something that used to require a publishing house, with automation handling the heavy mechanical lifting and the scarce expert attention reserved for the one thing only the experts can give.

Why it is possible now, and why it took these particular skills

So why now. That is the genuinely interesting part. Gathering this kind of knowledge has always been possible in theory and nearly impossible in practice, because the work was too large. To collect, structure, quality-assure, and keep alive an entire profession’s practical experience demanded an editorial operation no one could justify funding for a small Norwegian audience. That is the equation the new tools change. Not because the machine knows the craft, it does not. But because it can do the heavy, tedious labor around the craft. It can draft, structure, suggest, reconcile, and keep order, so that the few people who actually hold the tacit knowledge can spend their time on the one thing only they can do, which is to judge whether it is right.

But the tools alone do not build this. What this project has made clear to me is that the hard part sits exactly at the seam between the technical and the human. Anyone can wire a model to a database. Far fewer can build a retrieval system disciplined enough to abstain, a revision model honest about what decays, an audit trail that survives scrutiny, and then have the restraint to put a human expert at the final gate rather than letting the machine speak unchecked.

And here my not being a veterinarian stops being a caveat and starts being the point. My contribution is not clinical knowledge, which I do not have and do not pretend to. It is the ability to build the machinery that lets the people who do have it reach further: the systems thinking, the engineering, the stubbornness about sources and safety. The tools lowered the cost of the labor. The skill is in knowing where to spend the labor, and where to refuse to, and in being honest enough about my own limits to build the expert’s judgment into the center rather than my own.

What I am taking with me

When I lift my eyes from the specifics, a few things stay with me that I think hold far beyond one profession and one country.

The first is that the real problems are rarely where we first look. For a long time I thought this was about information, about making knowledge available. It was that too. But underneath lay a deeper problem about loneliness, and about how judgment erodes when the community around it crumbles. The tool got better the moment I stopped building against the symptom and started building against the actual absence.

The second is that the tacit, local knowledge is the valuable thing, precisely because it cannot be copied easily. General veterinary medicine exists in abundance, in textbooks and international databases. What barely exists in one place is the Norwegian, practical, production-animal experience. It had to be earned over time, by people who have actually stood in the barn.

The things that are hard to copy, it turns out, are usually just the things that were hard to earn.

The third, and the one I hold most firmly, is a stance toward the new tools that is neither fear nor infatuation. The machines finally make the large, heavy work affordable for small groups with deep knowledge and few hands. That is a gift, and it is new. But the gift is not that the machine can think for us. The gift is that it can clear the table, so the few who truly know the craft get more room to do what only people can do, and so the person standing alone at night feels a little less alone.

I still do not know how far this reaches, or how much of a professional community a tool can really stand in for. Probably less than I would like. But I have come to believe that the best systems we build in this era are not the ones that pretend to take over for judgment. They are the ones with the sense to keep it company.